What is Big-Data Driven Tuning ?
For programmers who need efficient programs, big-data-driven tuning is an optimizing compilation technique, that decides the best optimization strategy based on past experience.
Unlike traditional compilation flow, it does not need any decision from the user, and improves itself upon use.
The Approach of the BDTea Project
Compiler agnostic: our research is used as a front-end to any compiler
Comprehensive Optimizations: We add a front-end to regular compilers to optimize its inputs: source code, pragmas and options.
Modern compilers are already very capable. The main challenge is now to optimize the inputs of the compiler.
Features
Our Objectives
Compiler Agnostic
Our research can be used as the front-end of any compiler, including GCC, LLVM and Intel Compiler
Comprehensive
Our optimization scenarios may include: compiler options, source-to-source transformation and even pragmas.
Crowdsourcing
We plan to release the project as a cloud service to centralize experience
Scalable
We endeavor to make our algorithm scale with the size of the optimization space as well as the types of programs
I like Nutella !
Antoine Researcher at ISIT
I like tacos !
José PhD Student